Qwen3.6-35B-A3B for coding agents. With published tool-call accuracy.
OpenAI-compatible API (FP8, served on vLLM). Point Cline, Roo Code or Kilo Code at it in under a minute.
Tool-call accuracy scorecard
| Provider | Tool-call accuracy | Latency |
|---|---|---|
| Us: Qwen3.6-35B-A3B FP8 on vLLM 0.32, RunPod H100 SXM | 93.3% | about half of DeepInfra's |
| DeepInfra (same model) | 91.7% | baseline |
First live test: 30 coding-agent tool-call cases (valid JSON, correct tool, schema-valid args, semantic match, no hallucinated tools, correct abstention). Methodology and raw results will be published with each re-run.
Pricing (prepaid credits, pay as you go)
$0.08
per 1M input tokens
$0.04
per 1M cached input tokens
$0.75
per 1M output tokens
Most competitors charge about $0.10 input / $0.95 output per 1M tokens.
Quickstart
Sign up, create an API key in the dashboard, then:
Cline
Settings (⚙️) → API Provider: "OpenAI Compatible" Base URL: https://api.104-236-98-29.sslip.io/v1 API Key: sk-qa-…your key… Model ID: qwen3.6-35b-a3b
Roo Code
Settings → API Provider: "OpenAI Compatible" Base URL: https://api.104-236-98-29.sslip.io/v1 API Key: sk-qa-…your key… Model: qwen3.6-35b-a3b (Roo Code uses native tool calling only, which this API supports.)
Kilo Code
Settings → Providers → Custom provider Provider ID: qwen-coding-api Display name: Qwen3.6 Coding API Provider API: OpenAI Compatible Base URL: https://api.104-236-98-29.sslip.io/v1 API key: sk-qa-…your key… Models: qwen3.6-35b-a3b (or click "Fetch models"; /v1/models is supported)
curl
curl https://api.104-236-98-29.sslip.io/v1/chat/completions \
-H "Authorization: Bearer $API_KEY" -H "Content-Type: application/json" \
-d '{"model":"qwen3.6-35b-a3b","messages":[{"role":"user","content":"hi"}]}'